docs/src/content/docs/features/Canvas/bounding-box.mdx
The generation bounding box is the rectangle that defines where a Canvas generation takes place. It controls both the area returned by the model and the portion of the Canvas used as image context for that generation.
When you click Invoke, the visible Canvas content inside the bounding box is composited and sent through the generation pipeline. Content outside the box is not included in that Canvas image input, even though it remains visible to you on the larger Canvas.
This makes placement important. If the model needs to continue a shape, match a texture, or preserve part of a subject, include enough of that content inside the bounding box to give the model useful context. Moving the box changes which part of the image the model can use for the next generation.
:::note The bounding box defines the generation area; it is not an inpaint mask. Use an Inpaint Mask when only selected parts inside the box should be regenerated. :::
Select the Bounding Box tool in the Canvas toolbar. You can then:
The bounding box dimensions determine the generated image's width and height. Invoke constrains these dimensions to values supported by the selected model.
:::tip[Tap & Hold C] Tap <kbd>C</kbd> to keep the Bounding Box tool selected. Hold <kbd>C</kbd> to use it temporarily, then release the key to return to your previous tool. The shortcut can be changed in the Hotkeys settings. :::
The bounding box is the final footprint of the generation on the Canvas. If Scale Before Processing is enabled, Invoke may process that area at a different resolution and then resize the result back to the bounding box dimensions.
For best results, choose a box large enough to include the visual context the model needs, while keeping its dimensions appropriate for the selected model and your available VRAM.